Researchers working on the Google Brain initiative have developed artificial intelligence able to create its own encryption measures. This new function of AI will be able to compensate for the vulnerabilities posed by security systems created by humans as they have the keys to unravel the whole network.
Researchers working with deep learning technology at the Google Brain neural network program have recently published a paper that explains in more detail how the AI is able to create its own encryption and what implications could it have for our internet security.
According to the published article, researchers taught two neural networks, named Alice and Bob, to communicate with each other. They were also instructed to keep the contents of their dialogue a secret from a third-party network nicknamed Eve. In their pursuit of maintaining the privacy of their messages while being under attack by Eve, Bob and Alice were able developed new encryption systems.
The researchers discovered that the Google Brain AI created strange cryptographic schemes which are unlike anything humans have developed. Over the course of their experiment, they observed that at the start Eve was able to crack a few encryptions but as time passed Bob and Alice became more proficient at encrypting than Eve at decrypting.
The Google Brain researchers started the experiment by telling Allice to send Bob a message encrypted by 16-digit key containing only zeroes and ones. This shared key was gradually developed in a more complex encryption systems. For this reason, Eve was only able to decrypt the first 7,000 messages.
Researchers think that the AI’s proclivity of being better at encryption and decryption will have a revolutionary impact in a lot of domains. They could make sense of metadata faster than any human and could facilitate traffic analysis, for example.
Unfortunately, because they encryptions used by the AI aren’t entirely known to the researchers, will limit their use in practical applications. This means that although Ai could communicate without us knowing, the technology is still far from being useful in this stage at least. It is very likely that we won’t ever be able to crack the AI devised encryption.
What do you think about this new AI capacity developed by the Google Brain project?
Image source: Google